Abstract:In recent years, deep learning has made outstanding achievements in image classification, recognition, segmentation and generation. And some breakthroughs have been made in the research of image inpainting based on deep learning.The existing algorithms workwell, but they cannot inference in real time.In order to realize fast and efficient image inpainting,optimization is carried out from three aspectsbased on gated convolution. Using the pyramid sample to optimize the dilated gating convolution layers and proposing a coarse-to-fine pyramid sampling network(PUNet), compared with the gating convolution network, PUNet has less computation and more parameters to learn characteristics, as well as integrating different depth characteristics. Proposing holistic，pair-wise，pixel-wise loss function to enhance the local and global consistency. Introducing knowledge distillation into image inpainting and designs a multi-level self-distillation method. Experiments show that PUNet achieves the similar performance to gated convolutional network with 22% inference time.

Alert Name:
Alerting to:
Authentication email will be sent to your email address in 24 hours
Frequency:
Email Message Format:	Plain text Graphical(HTML)

Complete the form below and we will recommend the selected titles to your friends on your behalf. * Indicates a required field.
Your name*:
Your email address*:
Recipient's name*:
Recipient's email address*:
(multiple recipient's names and email addresses should be separated with semicolons)
Your comments:	I thought you would find the page(s) useful.

Your name:
Your email address:
Recipient's name:
Recipient's email address:
(multiple recipient's names and email addresses should be separated with semicolons)
Your comments:	I thought you would find this page useful.

Disclaimer: This message was sent to your friend using the "Send it to a friend" facility on the Sciencepaper Online’ WWW site, http://www.paper.edu.cn/en. The Sciencepaper Online is not responsible for the content of this email, and anything said in this email does not necessarily reflect the Sciencepaper Online's views.


	1. Efficient Image Inpainting with Knowledge Distillation
	Cheng Chuxuan,Shen Qiwei,Wang Jing
	Computer Science and Technology 10 January 2020
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:In recent years, deep learning has made outstanding achievements in image classification, recognition, segmentation and generation. And some breakthroughs have been made in the research of image inpainting based on deep learning.The existing algorithms workwell, but they cannot inference in real time.In order to realize fast and efficient image inpainting,optimization is carried out from three aspectsbased on gated convolution. Using the pyramid sample to optimize the dilated gating convolution layers and proposing a coarse-to-fine pyramid sampling network(PUNet), compared with the gating convolution network, PUNet has less computation and more parameters to learn characteristics, as well as integrating different depth characteristics. Proposing holistic，pair-wise，pixel-wise loss function to enhance the local and global consistency. Introducing knowledge distillation into image inpainting and designs a multi-level self-distillation method. Experiments show that PUNet achieves the similar performance to gated convolutional network with 22% inference time.
	TO cite this article:Cheng Chuxuan,Shen Qiwei,Wang Jing. Efficient Image Inpainting with Knowledge Distillation[OL].[10 January 2020] http://en.paper.edu.cn/en_releasepaper/content/4750420


	2. Semi-supervised Non-negative Matrix FactorizationBased on Semi-tensor Product
	WANG Lin, LI Li-Xiang, PENG Hai-Peng, YANG Yi-Xian
	Computer Science and Technology 30 December 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:Non-negative matrix factorization (NMF) is an effective feature extraction method. And the traditional NMF requires that the number of columns in the basis matrix is equal to the number of rows in the coefficient matrix, which imposes a great limitation on its engineering applications. Furthermore, some data in the practical applications may carry label information. These require novel methods to break this limitation and consider the influence of label information at the same time. Based on this idea, this paper proposes the semi-supervised non-negative matrix factorization based on semi-tensor product (TSNMF). The proposed method not only makes full use of the known label information, but also breaks through the limitation of dimension matching constraint in the traditional NMF, which can save storage space and improve the operation speed of the TSNMF method. Moreover, We evaluate the classification performance of the TSNMF method through numerical experiments in ORL face database and JAFFE face database. The experimental results show that the proposed TSNMF method is superior to the semi-supervised non-negative matrix factorization (SNMF).
	TO cite this article:WANG Lin, LI Li-Xiang, PENG Hai-Peng, et al. Semi-supervised Non-negative Matrix FactorizationBased on Semi-tensor Product[OL].[30 December 2019] http://en.paper.edu.cn/en_releasepaper/content/4750401

Saved Papers

Saved Papers


	3. Adaptive Parameters Softmax Loss for Deep Face Recognition
	ZHANG Jian-Wei,GUO Qiu-Shan,DONG Yuan,XIONG Feng-Ye,BAI Hong-Liang
	Computer Science and Technology 17 September 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:Face recognition has achieved great success due to the development of Deep convolutional neural networks (DCNN). Loss functions with angular margin have been proposed to supervise DCNN for better feature representation. However, these methods would suffer from sensitivity of hyperparameters setting. In this paper, we propose an Adaptive Parameters Softmax Loss function with different scale parameters for target logits and non-target logits and dynamically adaptive margin parameter. Extensive experiments on MegaFace and IJB-C demonstrate the effectiveness of our method.
	TO cite this article:ZHANG Jian-Wei,GUO Qiu-Shan,DONG Yuan, et al. Adaptive Parameters Softmax Loss for Deep Face Recognition[OL].[17 September 2019] http://en.paper.edu.cn/en_releasepaper/content/4749649


	4. Improved Face Super-Resolution GenerativeAdversarial Networks
	WANG Mengxue, Zhenxue Chen,Zhenxue Chen,ZHOU Xinjie
	Computer Science and Technology 28 June 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:The face super-resolution method is used for generating high-resolution images from low-resolution ones for better visualization. The Super-Resolution Generative Adversarial Network (SRGAN) can generate a single super-resolution image with realistic textures, which is a groundbreaking work. Based on SRGAN, we propose improved face super-resolution generative adversarial networks. The super-resolution image details generated by SRGAN usually have undesirable artifacts. To further improve visual quality, we delve into the key components of the SRGAN network architecture and improve each part to achieve a more powerful SRGAN. First, the SRGAN employs residual blocks as the core of the very deep generator network G. In this paper, we decide to employ Dense Convolutional Network blocks (Dense blocks), which connect each layer to every other layer in a feed-forward fashion as our very deep generator networks. Moreover, in the past few years, generative adversarial networks (GANs) have been applied to solve various problems. Despite its superior performance, however, it is difficult to train. A simple and effective regularization method called spectral normalization GAN (SNGAN) is used to solve this problem. We have experimentally confirmed that our proposed method is superior to the other existing method in training stability and visual improvements.
	TO cite this article:WANG Mengxue, Zhenxue Chen,Zhenxue Chen, et al. Improved Face Super-Resolution GenerativeAdversarial Networks[OL].[28 June 2019] http://en.paper.edu.cn/en_releasepaper/content/4749261


	5. A Lightened Sphereface for Face Recognition
	ZHOU Xinjie,Zhenxue Chen,WANG Mengxue
	Computer Science and Technology 25 June 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:Convolution neural networks (CNN) have significantly promoted the development of face recognition technology. In order to achieve ultimate accuracy, CNN models tend to be deeper or multiple local facial patch ensembles, resulting in excessive amounts of calculation. This paper addresses these deep face recognition (FR) problems and studies a lightened deep learning framework under an open-set protocol to achieve a good classification effect and streamline the model itself. To this end, we improve the Sphereface that enables deep network to learn angularly discriminative features faster and more effectively. First, global average pooling (GAP) is introduced to replace the original fully connected layer, which greatly reduces the size of the model. Compared to the widely used fully connected layer, GAP can reduce the number of parameters and avoid overfitting. Then Network in Network (NIN) layers are added between convolution layers. These models are trained on the CASIA-WebFace dataset and evaluated on the LFW and YTF datasets, which show the superiority of lightened SphereFace (L-SphereFace) in FR tasks. At the same time, computational cost is reduced by over nine times in comparison with the released SphereFace model. The size of the model is also close to the original half.
	TO cite this article:ZHOU Xinjie,Zhenxue Chen,WANG Mengxue. A Lightened Sphereface for Face Recognition[OL].[25 June 2019] http://en.paper.edu.cn/en_releasepaper/content/4749157


	6. Multilevel LSTM for Action Recognition Based on Skeleton Sequence
	CHEN Yan-Ru, PAN Hua-Wei
	Computer Science and Technology 17 April 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:Skeleton-based human action recognition has a broad range of applications in human-computer interaction and intelligent monitoring, and human behavior can be represented by the trajectory of the skeleton joint. Long-term short-term memory (LSTM) networks exhibit outstanding performance in 3D human action recognition because they are capable of modeling dynamics and dependencies in sequential data. In this paper, we propose a skeleton-based multilevel LSTM network for action recognition. First, the data for each joint and parent joint is used as input to a fine-grained subnet based on the action link between them. Then the features of the upper body joint are merged into the upper body subnet, the features of the lower body are merged into the lower body subnet, and finally the features of the two subnets are structured and fused to achieve higher recognition accuracy. Experimental results on the public data set NTU RGB+D demonstrate the effectiveness of the proposed network.
	TO cite this article:CHEN Yan-Ru, PAN Hua-Wei. Multilevel LSTM for Action Recognition Based on Skeleton Sequence[OL].[17 April 2019] http://en.paper.edu.cn/en_releasepaper/content/4748531


	7. A Spatial-aware Tracker
	Li Zhiyong,Xiang Ximing,Nai Ke,Jiang Shilong
	Computer Science and Technology 01 April 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:In this paper, a novel spatial-aware tracker (SAT), which utilizes the Siamese network and multiple correlation filters, is proposed to deal with fast motion and model drift problem in visual tracking. Specifically, the Siamese network is first used by an adaptive spatial search strategy to detect the target object in larger search areas. An extended search patch is generated if the target position obtained by the Siamese network is far away from the previous target position. Then, multiple correlation filters perform detection operations on both the extended search patch and the original search patch. With the proposed spatial selection scheme, SAT can accurately track the target object in challenging tracking scenes. By taking advantage of the Siamese network and multiple correlation filters, the proposed SAT tracker can effectively deal with fast motion and model drift problems to achieve better tracking performance. Extensive experimental results demonstrate that the proposed SAT tracker performs superiorly against several state-of-the-art trackers on OTB-2015 tracking benchmark.
	TO cite this article:Li Zhiyong,Xiang Ximing,Nai Ke, et al. A Spatial-aware Tracker[OL].[ 1 April 2019] http://en.paper.edu.cn/en_releasepaper/content/4748217


	8. Reconstruction-based Robust Pavement Crack Detection
	LUO Ling,XU Guosheng,XU Guoai
	Computer Science and Technology 21 January 2019
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:Pavement crack detection is of great importance for road maintenance. It is still very challenging to establish a unified and robust framework to perform accurate crack extraction from images with cluttered background, various morphological differences and even with shadow influence. In this paper, an improved semantic segmentation model with reconstruction branch is proposed for crack detection. Based on normal segmentation network, a deep convolutional encoder-decoder network is built to learn the image reconstruction mapping. This reconstruction guided semantic segmentation is aimed at improving detection accuracy by introducing reconstruction difference between crack and normal areas. The experiments demonstrated that our algorithm outperforms the convolutional segmentation method on two public datasets.
	TO cite this article:LUO Ling,XU Guosheng,XU Guoai. Reconstruction-based Robust Pavement Crack Detection[OL].[21 January 2019] http://en.paper.edu.cn/en_releasepaper/content/4747079


	9. Feature Selection Method Guided by Attention Mechanism for Image Classification
	Zang Hao,Huang Yaping,Tian Mei,Tian Mei,Tian Mei,1,1,1,1,1,1,1,1
	Computer Science and Technology 24 July 2018
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:In recent years, significant progresses have been made in the field of deep learning, especially in visual image classification. The features learned automatically by feed-forward deep convolutional neural networks (CNNs) are important for image classification. However, there is not much research on the spatial relationship of images in despite of a huge amount of work on constructing different network structures to improve classification accuracy. Therefore, in this paper, we propose a method for classifying images with saliency information and background information. We demonstrate that both saliency features and background features have an important influence on image classification. We firstly obtain attention heat map and features from CNN network. Secondly, we separate the features into saliency features and background features inspired by attention heat map. Then, we adopt several pooling strategies to process saliency and background features. Finally, we classify image by training a SVM classifier. Especially, we get effective improvements in Calthech-256 with 78.15\% accuracy and PASCAL VOC 2012 with 84.1\% mAP, demonstrating the effectiveness of our proposed method.
	TO cite this article:Zang Hao,Huang Yaping,Tian Mei, et al. Feature Selection Method Guided by Attention Mechanism for Image Classification[OL].[24 July 2018] http://en.paper.edu.cn/en_releasepaper/content/4745721


	10. AN EFFICIENT DEEP LEARNING HASHING NEURAL NETWORK FOR MOBILE VISUAL
	Heng Qi,Liang Liu
	Computer Science and Technology 26 January 2018
	Show/Hide Abstract \| Cite this paper︱Full-text: PDF (0 B)

	Abstract:Mobile visual search applications are emerging that enable users to sense their surroundings with smart phones. However, because of the particular challenges of mobile visual search, achieving a high recognition bitrate has becomes a consistent target of previous related works. In this paper, we propose a few-parameter, low-latency, and high-accuracy deep hashing approach for constructing binary hash codes for mobile visual search. First, we exploit the architecture of the MobileNet model, which significantly decreases the latency of deep feature extraction by reducing the number of model parameters while maintaining accuracy. Second, we add a hash-like layer into MobileNet to train the model on labeled mobile visual data. Evaluations show that the proposed system can exceed state-of-the-art accuracy performance in terms of the MAP. More importantly, the memory consumption is much less than that of other deep learning models. The proposed method requires only 13 MB of memory for the neural network and achieves a MAP of 97.80% on the mobile location recognition dataset used for testing.
	TO cite this article:Heng Qi,Liang Liu. AN EFFICIENT DEEP LEARNING HASHING NEURAL NETWORK FOR MOBILE VISUAL[OL].[26 January 2018] http://en.paper.edu.cn/en_releasepaper/content/4743206

	Check out RSS, or use RSS reader to subscribe this item